ai-memory
An open-source AI programming memory-sharing tool written in Rust. It automatically records agent interactions via lifecycle hooks, using Markdown files stored in a Git repository as the memory substrate. The tool supports cross-tool and cross-device context sharing as well as structured task handoffs.
Timeline (last 90 days)
ai-memory is an open-source project written in Rust for sharing memory across multiple AI coding assistants.
The ai-memory project has been in existence for approximately four months and has reached 8,120 GitHub stars, growing at a rate of over a hundred stars per day.
The core memory store in ai-memory consists of Markdown files in a Git repository, with the database serving only as an index that can be rebuilt from the original files after deletion.
The recording, retrieval, and handoff functions under ai-memory's default path do not require an API Key, and retrieval relies on keyword and link matching without depending on large language models.
ai-memory features a hierarchical forgetting mechanism where frequently used memories decay slowly and infrequently accessed notes are compressed into summaries, but the original records are preserved in the Git version chain and can be restored.
ai-memory manages task handoffs through a protocol-based system where tasks carry types and ownership, and the claim operation is atomic to prevent two agents from competing for the same task.
ai-memory chooses keyword and link matching over semantic retrieval, sacrificing some semantic recall capability in exchange for offline availability, zero API costs, and portability compatible with the Git toolchain.
All Facts (6)
ai-memory is an open-source project written in Rust for sharing memory across multiple AI coding assistants.
50%UnverifiedThe core memory store in ai-memory consists of Markdown files in a Git repository, with the database serving only as an index that can be rebuilt from the original files after deletion.
50%UnverifiedThe recording, retrieval, and handoff functions under ai-memory's default path do not require an API Key, and retrieval relies on keyword and link matching without depending on large language models.
50%Unverifiedai-memory features a hierarchical forgetting mechanism where frequently used memories decay slowly and infrequently accessed notes are compressed into summaries, but the original records are preserved in the Git version chain and can be restored.
50%Unverifiedai-memory manages task handoffs through a protocol-based system where tasks carry types and ownership, and the claim operation is atomic to prevent two agents from competing for the same task.
50%Unverifiedai-memory chooses keyword and link matching over semantic retrieval, sacrificing some semantic recall capability in exchange for offline availability, zero API costs, and portability compatible with the Git toolchain.
50%